Model Predictive Dispatch in Electric Energy Systems with Intermittent Resources Le Xie and Marija D. Ilic {lx,milic}@ece.cmu.edu March 11, 2009

Outline • Motivation • Problem Formulation • Proposed Algorithm for Economic Dispatch with Intermittent Resources • Numerical Examples • Summary and Future Work

Motivation • Increasing presence of renewable energy resources which are – Environmentally attractive – Intermittent

• Reliability and efficiency concerns for renewable resources due to intermittency • Three major questions: Better Prediction of Intermittent Resources?

More efficient utilization of intermittent resources More reliable operation of intermittent resources

This paper

Problem Statement • Economic Dispatch (ED): given a mixture of energy resources, how to determine the output of individual energy resources so that (1) power supply always balances demand (2) total generation cost is minimized? Total Gen Cost Supply=E(Demand ) Gen Constraints Ramp rate constraints

Conventional Approach to ED • Supply the expected load with whatever produced by intermittent resources combined with other traditional power plants. 4

1.7

x 10

Expected load

10000

Expected load

1.6

__

1.4

Intermittent resources output

x 10

Expected load minus the intermittent resources output Net Load

1.3

6000 MW

MW

4

1.5

8000

1.5

=

4000

1

2000

1.2 1.1 0

Expected intermittent resources output

500 1000 Time (5 minutes sample)

1500

0 0

500 1000 Time (5 minutes sample)

1500

0.5 0

500 1000 Time (5 minutes sample)

Economic Dispatch (ED): Choose output levels from conventional power plants to meet the “net load” at minimum cost.

1500

Conventional Approach to ED • Pros: – easy to implement – computes a reasonably good selection of generator outputs in “old days” when renewable resources are almost negligible in power systems

• Cons: – No flexible utilization of intermittent resources – High cost of keeping expensive fast-start units on in order to balance the high volatility of intermittent resources

Proposed Approach: Concept • Actively control the output of available intermittent resources to follow the trend of time-varying loads. • By doing so, the need for expensive fast-start fossil fuel units is reduced. Part of the load following is done via intermittent renewable generation. • The technique for implementing this approach is called model predictive control (MPC).

Model Predictive Control: Concept • MPC is receding-horizon optimization based control. • At each step, a finite-horizon optimal control problem is solved but only one step is implemented. • MPC has many successful real-world applications. www.jfe-rd.co.jp/en/seigyo/img/figure04.gif

Proposed Approach: Algorithm Lˆ (k ) Pˆ max (k )

Predictive Model and MPC Optimizer

wind

max Pˆsolar (k )

u k* : Output vector of all

Electric Energy System

generators at time step k

• Predictive model of load and intermittent resources are necessary. • Optimization objective: minimize the total generation cost. • Horizon: 24 hours, with each step of 5 minutes.

Numerical Experiment

4

1.7

x 10

Expected load Expected load

1.6

MW

1.5 1.4 1.3 1.2 1.1 0

500 1000 Time (5 minutes sample)

1500

Compare the outcome of ED from both the conventional and proposed approaches.

Numerical Experiment Conventional cost over 1 year *

Proposed cost over the year

Difference

Relative Saving

$ 129.74 Million

$ 119.62 Million

$ 10.12 Million 7.8%

*: load data from New York Independent System Operator, available online at http://www.nyiso.com/public/market_data/load_data.jsp

Summary • Look-ahead model predictive dispatch of future energy system is proposed. • Combined with good short-term prediction of intermittent resource outputs, the proposed method can lower the total generation cost. • The proposed method provides a benchmark towards optimal percentage of wind generation for grid and for storage. • More intelligent utilization of intermittent resources can actively follow the load variation trend, thus lower the total generation cost.

Future Work • Scale issue: how to make this algorithm fast enough in large-scale system? • Multi-objective problem: how to generalize the algorithm to study the tradeoff between environmental and economic costs? • More realistic model: how to include more realistic factors (e.g. transmission constraints) into the predictive dispatch model?

Acknowledgement

Thank you! Questions are welcome to be sent to {lx,milic}@ece.cmu.edu or visit us at www.ece.cmu.edu/~eesg